Skip to content

v0.7.1: enforce fp32 and torch kernels for triangle attention and multiplicative on V100.

Choose a tag to compare

View all tags
@zhangyuxuann zhangyuxuann released this 12 Nov 08:38
· 75 commits to main since this release
v0.7.1
ff68d0b

What's Changed

  • Add a dtype parameter to the Protenix CLI for inference, enabling FP32 inference via the -d flag.
  • For inference on V100 GPUs, certain configurations are forcibly adjusted — for example, BF16 precision and unsupported optimized kernels are disabled by default.
Assets 2
Loading